How to Use
Prompts to try, and how to get good results.
Copy any of these into your connected assistant and adapt it to your environment. Ask the way you would ask a colleague. Your assistant works out which tools to use.
Most requests start by picking an org. When a prompt names a sandbox, such as "my UAT sandbox," your assistant lists your registered sandboxes and asks which one you mean. You can also name it up front.
Ask about your configuration
Find validation rules
"List my registered nCino sandbox orgs. In the UAT one, show me all active validation rules on
LLC_BI__Loan__c, including their error messages."
Explore the data model
"What fields are on the nCino Loan object? Include the picklist values for the loan stage field."
Review access to an object
"In my UAT sandbox, list the profiles and permission sets that give access to the Loan object, then show me the object permissions for the one our loan officers use."
Look up a release
"Show me what's in nCino release 2026_08, including its package versions."
Search known issues
"Search nCino known issues for document generation problems in 2026_08, including its patch and integration releases, and summarize the top three."
Make a change
These change configuration or records in a sandbox. They need the change tools (also called write tools), which are off by default, so ask your Customer Success Manager to turn them on. Approval before a change runs is set in your AI assistant, so make sure that is configured first. See Limitations.
See Limitations.
Turn a feature flag on
"In my UAT sandbox, list the nFORCE feature flags that are currently off. Turn on the one for copying loans, and show me the change before you make it."
Set a system property
"In my dev sandbox, list the nFORCE system property categories related to pricing, then show me the properties in Commercial Pricing. Change
Rate_Sheet_Pricing_Matrix_PolicytoQ4_Rate_Sheet, and show me the current value first."
Add a picklist value
"In my UAT sandbox, add Credit Union to the Lost To picklist on the Loan object and to the record types that use it. Show me the current values and exactly what you will change before you deploy it."
Copy access between test users
"In my dev sandbox, list the permission sets assigned to our test underwriter user, then assign any that our new test user doesn't already have. Show me the assignments before you make them."
Feature flag names and system property categories are specific to your configuration, and both are case-sensitive. Ask the assistant to list what exists before you change anything.
Turning on a feature flag changes the flag only. It does not run the Enable Feature step in Feature Management, so if a feature is normally turned on there, check Feature Management afterward.
A new picklist value isn't added to record types automatically. Ask for the record types you want, or add them in Setup.
Work across several steps
Plan an upgrade
"In my UAT sandbox, work out which nCino release we are on and build the upgrade path to the latest release. Then compare the
LLC_BI__Loan__cschema between the two releases, list any objects that reference Loan, and flag anything that could break our validation rules."
Audit configuration
"In my UAT sandbox, find every active record-triggered flow on the Loan object, list the validation rules that could conflict with them, and check the debug logs from the last 30 minutes for related errors."
Test Apex before a release
"In my dev sandbox, find our own Apex tests that exercise the Loan object, run them, and show me any failures and which of our classes are under 75% coverage."
Check the docs against reality
"Search nCino technical documentation for how to configure Loan Auto Decision, then check whether the flows, permission sets and system properties it describes already exist in my UAT sandbox."
Debug logs only exist for users with an active trace flag, which you set in Setup. Apex tests run in the background and can take several minutes. Only your own (unmanaged) tests can run, not nCino's packaged ones.
Getting good results
- Say which org. The connector never guesses. Naming it up front saves a round trip.
- Ask for a preview before a change. Ask your assistant to show you the current value and what it intends to do.
- Check the result after a change. Batched changes are not all-or-nothing, so when a change matters, ask the assistant to read the values back afterward.
- Chain steps, but keep the scope tight. The assistant can carry out several steps in one request, and asking for the whole job usually works better than three separate questions. What to avoid is breadth. A question about one object or one release comes back complete, while one spanning everything at once may come back shortened.
Updated 9 days ago
